QUADRO FOR EDUCATION & QUADRO FOR EDUCATION & RESEARCH - - PowerPoint PPT Presentation

▶
quadro for education quadro for education research
SMART_READER_LITE
LIVE PREVIEW

QUADRO FOR EDUCATION & QUADRO FOR EDUCATION & RESEARCH - - PowerPoint PPT Presentation

Dec 18, 2022 228 likes •385 views

QUADRO FOR EDUCATION & QUADRO FOR EDUCATION & RESEARCH RESEARCH 6-2019 EDUCATION ENVIRONMENT CHALLENGES Applications. Frameworks, Libraries Research problems, Educational and Demand for skilled data IT budgets not growing

slide-1
SLIDE 1

6-2019

QUADRO FOR EDUCATION & QUADRO FOR EDUCATION & RESEARCH RESEARCH

slide-2
SLIDE 2

2

EDUCATION ENVIRONMENT CHALLENGES

Research problems, student projects more complex than ever, data sets & compute requirements growing exponentially Educational and research software toolchains & workloads increasing in size and complexity Demand for skilled data scientists & AI expertise requires students to hit the ground running after graduation IT budgets not growing to match technology infrastructure demands

Applications. Frameworks, Libraries

slide-3
SLIDE 3

3

QUADRO HER VALUE PROPOSITION

Only NVIDIA Quadro solutions provide:

  • Compute power, GPU memory capacity, and

scalability required for today’s demanding education & research projects

  • Researcher and student access to

professional level hardware & software – same used by professionals around the world

  • Enterprise level hardware & software

support – maximize uptime and minimize IT support requirements

  • Server ready GPU solutions

RTX 6000 24GB

slide-4
SLIDE 4

4

QUADRO FOR DEEP LEARNING

Quadro value for DL training:

  • Large memory, scalability for large

training data sets, model optimization & validation

  • Scalability with NVLink to speed up

training tasks, provide development platform for server deployments

TRAINING INFERENCE

Quadro value for DL inferencing:

  • Fast inferencing performance for the

largest networks & datasets

  • Support for encode/decode of

multiple simultaneous video streams

  • Consumer cards limited to 2 simultaneous

streams

Quadro solutions are validated, tested, and used daily in demanding professional DL environments

slide-5
SLIDE 5

5

DEEP LEARNING TRAINING PERFORMANCE

Quadro Provides Memory & Performance Required for DL Training

Large training datasets such as natural language processing, machine translations, etc., don’t fit into consumer graphics memory Quadro enables larger batch sizes to accelerate DL training *

*Did not run due to insufficient memory

1.36 1.69 RTX 2080 RTX 2080Ti RTX 6000 0.0 0.2 0.4 0.6 0.8 1.0 1.2 1.4 1.6 1.8

ResNet-50 Training Mixed Precision

Relative Performance

0.00 2.03 RTX 2080 RTX 2080Ti RTX 6000 0.00 0.50 1.00 1.50 2.00 2.50

OpenSeq2Seq (GNMT) Training Mixed Precision

Relative Performance

Test run on Xeon Gold 6140@2.30GHz, NVIDIA driver 415.13, Tensor Flow ResNet-50 and OpenSeq2Seq training benchmarks, mixed precision.

slide-6
SLIDE 6

6

DEEP LEARNING INFERENCING PERFORMANCE

Deep Networks with Many Layers can Take Advantage of Quadro Memory

1.29 1.33 RTX 2080 RTX 2080Ti RTX 6000 0.00 0.20 0.40 0.60 0.80 1.00 1.20 1.40

VGG16 Inference Mixed Precision

Relative Performance

1.25 1.31 RTX 2080 RTX 2080Ti RTX 6000 0.00 0.20 0.40 0.60 0.80 1.00 1.20 1.40

VGG19 Inference INT8 Precision

Relative Performance

Test run on Xeon Gold 6140@2.30GHz, NVIDIA driver 415.13, Tensor Flow VGG16 and VGG19 inference benchmarks, mixed precision.

slide-7
SLIDE 7

7

QUADRO – FASTER TIME TO SOLUTION*

40.2 60 50 40 30 20 10 RTX 2080 Ti RTX 6000

1 Hourt Training Time (in minutes)

16.1 24 20 16 12 8 4 RTX 2080 Ti RTX 6000

1 Day Training Time (in Hours)

4.7 7 6 5 4 3 2 1 RTX 2080 Ti RTX 6000

1 Week Training Time (in Days)

*based on RTX 6000 ResNet-50 performance 33% faster than 2080Ti

slide-8
SLIDE 8

8

QUADRO REQUIRED FOR THE LARGEST WORKLOADS

Data Science Data Sets Require GPU Memory Only Available with Quadro

2080Ti RTX 6000 1 2 3 4 5 6 7

Max Data Set Size

(in months)

Quadro value for data science

  • Large Quadro memory lets

data scientists process more data to improve model training and accuracy.

  • Quadro performance

completes tasks faster Sample data set using Home Mortgage data in the US for 2016. A single GeForce RTX 2080Ti can only load 3 months worth of data. A single Quadro RTX 6000 GPU can load 6 months of data at a time. Two RTX 6000’s can load the entire years worth of data in the combined GPU memory (dual 2080Ti’s can only load 6 months of data).*

1.40 2080Ti RTX 6000 0.00 0.20 0.40 0.60 0.80 1.00 1.20 1.40 1.60

End To End Training Time

(3 months of data, relative performance)

Data Science Workload Example:

*A single RTX 8000 can load the entire year’s worth of data

Test system: dual Gold 6140@2.30GHz 3.7GHz Turbo (Skylake), ETL with Dask + Pandas

slide-9
SLIDE 9

9

QUADRO FOR HPC

Quadro value for HPC:

  • Larger GPU memory for the largest data

sets and compute tasks

  • Memory and compute power enables

multi application workflows from compute to visualization

  • Quadro is NGC Ready, tested and

validated for NVIDIA NGC HPC software

slide-10
SLIDE 10

10

QUADRO FOR VIRTUALIZATION

Quadro value for virtualization:

  • Get the features, performance of Quadro

anywhere you need it.

  • Partition a single RTX 6000 into multiple high

performance virtual workstations

  • Quadro vDWS only available on Tesla, Quadro

RTX 6000/8000 GPUs Learn more about Quadro vDWS solutions:

https://www.nvidia.com/quadro-vdws/

Quadro vDWS – Virtual Quadro Anywhere

slide-11
SLIDE 11

11

NVIDIA NGC SUPPORT SERVICES

Minimize Downtime And Maximize System Utilization

Availability

  • Exclusively for NGC-Ready

workstations

  • Service agreement between

NVIDIA & customer

  • Purchase from OEM/System

vendor Support by NVIDIA’s subject matter experts 24x7 portal, phone and email access to create support cases Live support during local, regional business hours for technical assistance

Support Coverage

  • NGC DL & ML containers
  • NVIDIA drivers
  • NV-docker
  • CUDA
slide-12
SLIDE 12

12

NVIDIA Desktop System Use Case

PC-based HPC, AI/DL/ML workloads, small to medium size data sets, up to 22GB memory w/NVLink Workstation-based HPC, AI/DL/ML workloads, largest data sets with up to 48GB memory w/NVLink, pre-installed data science software stack.

GPU Memory

11GB 24GB

ECC Memory No Yes – error free compute NVLink support 2-way NVLink, 3 & 4 slot bridges 2-way NVLink, 2 & 3 slot bridges – fits into a wide variety of workstation chassis Cooling Solution Dual Axial Cooling Blower Active Fan – ideal cooling solution for multi-GPU configurations Pre-installed Data Science Software Stack from OEMs No Yes – up and running data science workloads within minutes Nvidia support N/A Enterprise level pre/post sales support OEM support N/A Enterprise level support NGC Ready No Yes – validated and tested for running NVIDIA NGC software NVIDIA NGC Support Services Option N/A Yes – support provided directly from NVIDIA Quadro Advantage

฀ Work with largest data sets to accelerate HPC, AI, ML workflows. NVIDIA and OEMs provide enterprise level testing, validation, and support. Additional NVIDIA NGC Support Services availability. GPU Cooling & NVLink solutions support widest range

  • f system configurations.

Quadro RTX 6000 GeForce RTX 2080Ti

slide-13
SLIDE 13

13

QUADRO FOR HIGHER EDUCATION & RESEARCH

  • Enterprise grade hardware, enterprise level

software & hardware testing, validation, & support – ready for workstation & server deployments

  • Performance & scalability – large GPU memory

for large datasets & workloads today and room for growth in the future

  • Maximum flexibility with the option to provide

multiple virtual Quadro workstations (Quadro vDWS) from a single Quadro GPU

Quadro Brings Real Value

RTX 6000 24GB

slide-14
SLIDE 14